QuantFactory/bigstral-12b-32k-GGUF
Full fine-tune (fewer layers) vs mistralai/Mistral-7B-Instruct-v0.2. mistralai/Mistral-7B-Instruct-v0.2 matches part of QuantFactory/bigstral-12b-32k-GGUF (291 of its 507 tensors; the rest of QuantFactory/bigstral-12b-32k-GGUF has no counterpart). On the shared part: QuantFactory/bigstral-12b-32k-GGUF is a GGUF (Q8_0) of different weights. Fine-tune: 100% of the weights changed, across 32 of 32 layers.
Tensors changed: Embeddings 1/1, Norms 48/65, MLP 96/96, Attention 128/128, lm_head 1/1